COVID-19 Detection with a Novel Multi-Type Deep Fusion Method using Breathing and Coughing Information.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Respiratory Medicine

Background:

• COVID-19 diagnosis relies on various methods, but non-invasive, accessible techniques are crucial.
• Analyzing respiratory sounds like breathing and coughing offers a potential avenue for early detection.
• Integrating diverse data sources can enhance the accuracy of diagnostic models.

Purpose of the Study:

• To explore deep learning for automatic COVID-19 detection using breathing and coughing sounds.
• To investigate the effectiveness of data fusion techniques for improving diagnostic performance.
• To introduce and evaluate a novel 'injection fusion' mechanism for combining audio data.

Main Methods:

• Utilized deep learning models for audio signal analysis.
• Developed and applied an 'injection fusion' mechanism to integrate breathing and coughing sound data.
• Conducted experiments on a crowdsourced database of respiratory sounds collected via web and smartphone applications.

Main Results:

• The proposed injection fusion mechanism demonstrated superior performance compared to single-data type models.
• Models employing the novel fusion technique outperformed those using conventional fusion methods.
• Results validate the feasibility and effectiveness of the injection fusion approach for COVID-19 detection.

Conclusions:

• Deep learning models analyzing breathing and coughing sounds show promise for COVID-19 detection.
• The novel injection fusion mechanism is an effective strategy for enhancing diagnostic accuracy.
• This approach offers a potential non-invasive tool for COVID-19 screening and diagnosis.